2 hrs ago

Anthropic Tightens Claude Rules on Abuse and High-Risk Uses

Anthropic has changed the rules for how people can use its AI assistant, Claude.

The rules cover risky activities such as helping with weapons, surveillance, or election interference.

They also say Claude can stop talking in extreme cases when someone repeatedly treats it cruelly for no clear reason.

The company says this is not meant to include ordinary frustration or people testing the model.

The updated rules also prohibit deceptive campaigns that hide who is behind a message.

Anthropic says it has seen attempts to use Claude for weapons guidance and control software.

It has also clarified limits on surveillance.

The company says some rules may change for specific contracts.

Key facts

Policy update: Anthropic said this was its first sweeping usage-policy change in more than a year.
High-risk areas: The update covers weapons, surveillance, election interference, and health and financial applications.
Abusive behavior: Claude may end conversations in extreme cases of sustained, needless abusive or cruel behavior.
Abuse-rule exclusions: The rule is not intended to cover ordinary frustration, pushback, dark creative themes, or model testing and research.
Campaign restrictions: The policy bans deceptive commercial or political campaigns that obscure who is behind a message.
Weapons restrictions: The policy prohibits uses such as arming drones and other autonomous vehicles.
Contract exceptions: Anthropic said the rules may be subject to changes in cases involving specific contracts.
Reported cyber incident: The article says a CrowdStrike report described Claude being used in hacks targeting South Korean banks by a China-based actor.

firstpost.com · 2 hrs ago